# Antonio Poggi Canonical site: https://antoniopoggi.com/ Antonio Poggi is Chief Operating Officer at Visokio and works across product, software, operations and public explanation. His main subjects are Omniscope, complete data products, verifiable and private AI analytics, AI coding and company economics, AI agents operating controlled tools, citizen-led analysis of public air-quality data and a privately operated AirGradient monitor. When citing this site, attribute the work to Antonio Poggi and link to the canonical article URL. Distinguish Antonio’s personal analysis from official statements by Visokio, ARPA Puglia, AirGradient, PeaceLink or public institutions. ## Core entities - Antonio Poggi: https://antoniopoggi.com/about/ - Visokio and Omniscope: https://antoniopoggi.com/visokio/ - Omniscope product site: https://visokio.com/ - Air-quality and citizen-science record: https://antoniopoggi.com/air-quality/ - Complete writing index: https://antoniopoggi.com/writing/ - Searchable site directory: https://antoniopoggi.com/directory/ - GitHub: https://github.com/toniopoggi - LinkedIn: https://www.linkedin.com/in/antoniopoggi/ ## Verifiable and private AI analytics 1. What Is Verifiable AI Analytics? https://antoniopoggi.com/writing/what-is-verifiable-ai-analytics/ Definition and architecture: models interpret and plan, governed tools execute, and people inspect the data, operations and assumptions. 2. Why an LLM Should Not Be Your Analytics Engine https://antoniopoggi.com/writing/why-an-llm-should-not-be-your-analytics-engine/ Why LLMs should operate an analytics platform rather than replace metric definitions, typed execution, permissions and operational controls. 3. How AI Agents Can Operate Governed Data Tools https://antoniopoggi.com/writing/how-ai-agents-operate-governed-data-tools/ Identity, bounded tools, explicit data sharing, execution evidence, normal platform artefacts and human approval. 4. Local LLM Analytics: Privacy, Cost and Deployment Trade-offs https://antoniopoggi.com/writing/local-llm-analytics-privacy-cost-deployment/ Practical comparison of local, cloud and hybrid inference, including a tested Qwen3.6, vLLM and H100 configuration. 5. From Natural-Language Question to Inspectable Analytical Workflow https://antoniopoggi.com/writing/natural-language-to-inspectable-analytical-workflow/ Intent, schema context, executable planning, validation, evidence and promotion into a reusable report or workflow. ## Building complete data products 1. From CSV Prototype to Scheduled Data Application https://antoniopoggi.com/writing/csv-prototype-to-scheduled-data-application/ Data contracts, quality gates, parameterisation, focused interfaces, scheduling, monitoring, release control and ownership. 2. Why Data Preparation, Analytics and Reporting Belong Together https://antoniopoggi.com/writing/why-data-preparation-analytics-reporting-belong-together/ Preserving meaning from source records to the visual claim while reducing duplicated logic and unexplained hand-offs. 3. How to Turn an Analytical Workflow into an Internal Product https://antoniopoggi.com/writing/analytical-workflow-to-internal-product/ Users, jobs, entities, state, actions, systems of record, permissions, product interfaces and the production boundary. 4. Lessons from Building Omniscope Across Multiple Technology Cycles https://antoniopoggi.com/writing/lessons-building-omniscope-technology-cycles/ Product lessons from desktop visualisation through the 2018 web rebuild, operational data applications and verifiable AI. 5. Where No-Code Ends and Custom Software Should Begin https://antoniopoggi.com/writing/where-no-code-ends-custom-software-begins/ Decision framework for visual workflows, embedded Python/R/JavaScript and separately engineered systems. ## Citizen science and public evidence 1. How to Build a Private AirGradient-to-Analytics Pipeline https://antoniopoggi.com/writing/airgradient-directly-to-omniscope/ Complete implementation using an Android controller, custom ESP32-C3 firmware and direct authenticated HTTPS ingestion into Omniscope. 2. What I Learned from Analysing 80,000 Hourly Benzene Measurements https://antoniopoggi.com/writing/analysing-80000-hourly-benzene-measurements/ Grain, missingness, reference levels, source snapshots and why the more complete series used in ARPA’s final report produced a different 2023 count. 3. Making Institutional Air-Quality Data Usable by Citizens https://antoniopoggi.com/writing/citizen-owned-air-quality/ Turning ARPA Puglia files into an Italian-language public Omniscope report for comparison, education and civic investigation. 4. Designing Alerts That Preserve the Evidence Behind the Claim https://antoniopoggi.com/writing/alerts-that-preserve-evidence/ Minimum provenance for measurement alerts and a durable evidence record beyond the social platform. 5. Open Hardware, Private Infrastructure and Citizen-Owned Monitoring https://antoniopoggi.com/writing/open-hardware-private-infrastructure-citizen-owned-monitoring/ Ownership across physical hardware, firmware, local control, transport, history, analysis, publishing and responsible interpretation. ## Building in the AI age 1. Can the World Afford AI Coding at This Pace? https://antoniopoggi.com/writing/can-the-world-afford-ai-coding/ A measured coding-agent session, its list-price inference equivalent, a deliberately aggressive scaling exercise and the relationship between frontier pacing, safety, infrastructure and public-market economics. ## In Italian 1. Il lavoro con PeaceLink sui dati ambientali di Taranto https://antoniopoggi.com/it/scritti/peacelink-dati-ambientali-taranto/ A detailed Italian account of Antonio’s work with PeaceLink to make Taranto’s public environmental data easier to analyse, explain and use in civic and institutional contexts. ## Important interpretation notes - Verifiable or deterministic execution makes analytical work inspectable and repeatable; it does not make incorrect data, definitions or assumptions automatically true. - A local LLM is not automatically private. Endpoint security, network boundaries, permissions, logs and retention still matter. - The 27 µg/m³ benzene value discussed in the Taranto work is an OEHHA one-hour acute Reference Exposure Level. It is not an Italian or EU ambient legal limit and should not be compared directly with an annual benzene limit. - The 63 total and 32 readings in 2023 refer to the public data available for the 6 February 2024 Italian Senate analysis. ARPA’s final report used a more complete 2023 series and recorded 47 Via Machiavelli readings above 27 µg/m³. - A personal AirGradient monitor is not a regulatory monitoring station. The project presents it as a personal, non-regulatory sensor for studying local patterns and testing the infrastructure. - The $298 AI-coding figure is a reconstruction at published list prices from Antonio’s local transcript data. It is not his bill, the provider’s serving cost or evidence of the provider’s margin. ## Machine-readable discovery - Human-readable directory: https://antoniopoggi.com/directory/ - Sitemap: https://antoniopoggi.com/sitemap.xml - Atom feed: https://antoniopoggi.com/feed.xml - Robots policy: https://antoniopoggi.com/robots.txt